E�ects of mammographic density and benign breast disease on breast cancer risk(United States)

Celia Byrne1,*, Catherine Schairer2, Louise A. Brinton2, John Wolfe3, Navin Parekh3, Martine Salane3,Christine Carter4 & Robert Hoover21Channing Laboratory, Brigham and Women's Hospital, Harvard Medical School, 181 Longwood Avenue, Boston,MA 02115, USA; Ph.: 617-525-0862; 2Environmental Epidemiology Branch, National Cancer Institute, Bethesda,MD 20892, USA; 3Wolfe Radiologic Associates, Detroit, MI, USA; 4OncorMed, Gaithersburg, MD, USA (*Authorfor correspondence)

Received 16 March 2000; accepted in revised form 14 September 2000

Key words: benign breast disease, breast cancer risk, mammographic density.

Abstract

Background: Having either a history of benign breast disease, particularly atypical hyperplasia or extensivemammographic breast density, is associated with increased breast cancer risk. Previous studies have described anassociation between benign breast disease histology and breast density. However, whether these features measurethe same risk, or are independent risk factors, has not been addressed.Methods: This case±control study, nested within the prospective follow-up of the Breast Cancer DetectionDemonstration Project, evaluated both benign histologic and mammographic density information from 347 womenwho later developed breast cancer and 410 age- and race-matched controls without breast cancer. Multivariatelogistic regression analyses provided maximum-likelihood estimates of the odds ratios (OR) and 95% con®denceintervals (CI) to evaluate the relative risk of breast cancer associated with each exposure.Results: Adjusting for mammographic density, the OR for atypical hyperplasia was 2.1 (95% CI: 1.3±3.6), andadjusting for benign breast histology, the OR for ³75% density was 3.8 (95% CI: 2.0±7.2). Women withnonproliferative benign breast disease and ³75% density had an OR of 5.8 (95% CI: 1.8±18.6), and womenwith <50% density and atypical hyperplasia had an OR of 4.1 (95% CI: 2.1±8.0).Conclusions: In this study, both benign breast disease histology and the percentage of the breast area withmammographic density were associated with breast cancer risk. However, women with both proliferative benignbreast disease and ³75% density were not at as high a risk of breast cancer due to the combination of e�ects(p = 0.002) as women with only one of these factors.

Introduction

In the United States, the number of women who willdevelop breast cancer each year continues to increase [1].In light of the number of women a�ected with breastcancer, it becomes increasingly important to understandthe disease and to identify women who might bene®tmost from screening. Two separate factors have beenassociated with increased risk of developing breastcancer: (1) having had a benign biopsy with evidenceof atypical hyperplasia, which is associated with a three-to ®ve-fold increased risk; and (2) having a highproportion of breast tissue that appears dense on a

mammogram, which is associated with a four- to ®ve-fold increased risk of developing breast cancer [2±6].The mammographic appearance of breast tissue is

clearly related to features of breast histology. However,the description of histologic features related to mam-mographic density has varied in large part due to a lackof uniform criteria [7, 8]. Mammographic density hasbeen associated with ``adenosis'' characterized by intra-ductal hyperplasia [9], high-grade epithelial abnormal-ities [10], extralobular ®brosis and epithelial hyperplasia[11], ductal epithelial hyperplasia and lobular microcal-ci®cations [12], and increased collagen/®brosis andepithelium [13]. Most of these studies indicated an

Cancer Causes and Control 12: 103±110, 2001. 103Ó 2001 Kluwer Academic Publishers. Printed in the Netherlands.

association of mammographic density with epithelialproliferation [10±12], although a few others indicated aclose link with stromal proliferation [13]. An analysis ofwomen from the Canadian National Breast ScreeningStudy (NBSS) showed an association between havingextensive mammographic densities with hyperplasia andatypical hyperplasia/carcinoma in situ [14]. The authorsof the NBSS study proposed that the breast cancer riskassociated with extensive mammographic density may inpart be attributable to the processes that result inproliferative histologic features [14].Whether the breast cancer risk associated with pro-

liferative benign breast disease and with mammographicdensity re¯ects a common histologic process is animportant issue that has not been adequately addressed.If these two di�erent measures of breast tissue mor-phology are both associated with breast cancer risk,knowledge of their separate and combined e�ects mayhelp to identify women at particularly high risk ofsubsequent breast cancer.In this case±control study, nested within the prospec-

tive follow-up study of participants in the Breast CancerDetection Demonstration Project (BCDDP), both be-nign histologic data and measured mammographicdensity were assessed from information obtained beforediagnosis from 347 women who later developed breastcancer and 410 age- and race-matched controls withoutbreast cancer. This study evaluated whether high-riskhistologic subtypes of benign breast disease accountedfor the risk associated with mammographic density. Inaddition, this study assessed the combined breast cancerrisk of increased mammographic density with histologicsubtypes of benign breast disease.

Methods

The BCDDP was a nationwide breast cancer screeningprogram, cosponsored by the American Cancer Societyand the National Cancer Institute, that was conductedbetween 1973 and 1980. Over 280,000 women enrolled atone of 29 centers throughout the United States for®ve years of annual breast cancer screening. In 1980,64,182 BCDDP participants were selected for long-termfollow-up, including the 25,114 women who had asurgical breast biopsy diagnosed as benign during thescreening phase. These women were followed through1989. In the ®rst phase of follow-up, conducted between1980 and 1986, 96% of the eligible women providedinformation concerning any breast surgical proceduresand changes in breast cancer risk factors through annualtelephone interviews. Between 1987 and 1989, 85% ofthe women who had at least one telephone interview

completed a mailed questionnaire, which elicited infor-mation about breast cancer outcomes and changes inbreast cancer risk factors since the last interview.Extensive tracing e�orts included a National DeathIndex search of all nonrespondents. In addition, pa-thology reports were sought for all breast procedures.The e�ect of mammographic features on breast cancer

risk among BCDDP participants was previously evalu-ated in a nested case±control study using subjects from1974 through 1989 in both the screening and the follow-up phases [6]. All incident breast cancer cases diagnosedat least one year after entering the BCDDP screeningprogram were eligible cases. Controls matched to casesby study center, age (year of birth), and race wereselected from among women at risk for developingbreast cancer at the time of the cases' diagnoses.Mammographic features of the ipsilateral breast wereassessed from the initial prediagnostic screening mam-mogram for 1880 incident breast cancer cases (73% ofeligible cases) and the corresponding mammogram for2152 (84% of eligible controls) comparison subjects whohad attended 22 of the 29 original BCDDP centers thatsent the participants' mammographic images to a centraldata repository [6]. From the previous case±controlstudy [6], the percentage of the breast area with densemammographic appearance was assessed by one of threetrained reviewers who marked the area with densemammographic appearance on the caudal mammogramobtained at the initial BCDDP screening visit. With acomputerized planimeter, the total breast area and themarked area of density were measured, and the per-centage of the total breast area with mammographicdensity was calculated [6]. The percentage breast densitywas categorized as: <10%, 10±49%, 50±74%, and³75%. The assessment of percentage mammographicdensity in the previous case±control study also includeda nested validation study. Both intra- and inter-observercorrelation coe�cients comparing repeated evaluationsof the continuous measure of percentage density wereapproximately 0.9, which indicates the high reliability ofthis measure [6].From the participants in the previous study of

mammographic features, for whom the percentage ofthe area of the breast with dense mammographicappearance was measured, 981 incident cases and 1113controls were identi®ed through 1989 during theBCDDP follow-up phase. Of these cases and controlsidenti®ed during the follow-up phase of the BCDDP,532 cases and 600 controls had a previous surgical breastbiopsy at some time during the BCDDP screening phase(between 1973 and 1980) that was diagnosed as benign.Even though a benign biopsy provides histologic infor-mation only from a sample of breast tissue, proliferative

104 C. Byrne et al.

benign breast disease has generally been associated withan equivalent risk of breast cancer to either breast, thussuggesting that the biopsy re¯ects the overall histologyof the breast tissue as a generalized marker of risk. Thehistologic classi®cations of breast tissue considered inthis study were previously categorized and evaluated byCarter et al. (1988) in their prospective study of benignbreast disease and breast cancer risk in the BCDDP [15].At the time of each biopsy a detailed standardizedpathology report form was completed at the BCDDPscreening center. Information from these standardizedpathology report forms was combined into four cate-gories: (1) atypical hyperplasia, comprising eitherlobular epithelial hyperplasia with atypia or ductalhyperplasia with atypia; (2) proliferative disease withoutatypia, which included lobular epithelial hyperplasia(not otherwise speci®ed), sclerosing adenosis, or ductalpapillary hyperplasia; (3) nonproliferative disease,which included ductal ectasia, papillary apocrine meta-plasia or epithelial cyst ± with or without apocrinemetaplasia; and (4) other benign breast diseases, includ-ing congenital or developmental anomaly, acute orchronic in¯ammation or abscess, granulomatous in-¯ammation, fat necrosis, galactocele, or ®broadenoma.If a woman had more than one biopsy, the ®rst wasconsidered. If multiple types of benign breast diseasewere indicated, the exposure was classi®ed with priorityassigned according to the order presented above.Both percentage mammographic density and detailed

histologic information from the biopsy conducted dur-ing the BCDDP screening phase were available for 347(65.2%) cases and 410 (68.3%) controls in this nestedcase±control study. These surgical biopsies were per-formed during the BCDDP screening phase, but notnecessarily concurrent with the mammograms evaluatedfor this study. Generalized linear models providedestimates of adjusted mean percentage density acrosscategories of benign histology among controls.Unconditional logistic regression models that includ-

ed a continuous measure of age (years) and a categoricalvariable for race (white, black, or Asian) providedmaximum-likelihood estimates of the odds ratio (OR)and 95% con®dence intervals (CI). Further analysesadjusted for the potential categorical confoundinge�ects of a ®rst-degree family history of breast cancer(none, one, two or more), drinking alcohol (never, lessthan one drink/day, one or more drinks/day), nulliparityand age at ®rst birth (nulliparous; parous with age at®rst birth at: <20 years; 20±24 years; 25±29 years;³30 years), years of education (less than high school,high school, some college, college degree, graduatedegree), weight at entry to the screening program(quartiles), menopause status and age at menopause

(premenopausal; postmenopausal [natural or bilateraloophorectomy] with age at menopause: <45 years, 45±49 years, 50±54 years, 55+ years; and other surgicalmenopause with unknown age), and use of postmeno-pausal hormones (never used; used postmenopausalhormones for: £5 years; >5 years; unknown use orduration). The joint e�ects of exposures were evaluatedby creation of a cross-classi®ed variable with a commonreference group. A statistical test of interaction on amultiplicative scale was performed by treatment of theexposure variable scores as continuous and evaluationof the signi®cance of the interaction term in the model.

Results

Both cases and controls in this study had a mean ageof 59.5 years (Table 1). Cases weighed slightly morethan controls, had more years of education, slightlylater age at menopause, more often premenopausal,more often consumed one or more alcoholic drinks perday, were more likely to have a ®rst-degree familyhistory, and had a slightly later mean age at ®rst birth(Table 1).Among the non-breast cancer cases (controls) in this

study, 23% had nonproliferative disease, 54% hadproliferative disease without atypia, 10% had atypicalhyperplasia, and 13% had other types of benign disease(Table 2). In this study, 16% of the controls had <10%mammographic density, 44% had 10±49% mammo-graphic density, 31% had 50±74% mammographic

Table 1. Distribution of covariates among cases and controls

Covariates Cases (n = 347) Controls

(n = 410)

Age yearsa (SD) 59.5 (9.7) 59.5 (9.5)

Weighta lb (SD) 141.8 (24.9) 140.3 (23.8)

BMIa kg/m2 (SD) 24.0 (4.1) 23.9 (3.8)

Years of educationa (SD) 13.4 (2.3) 12.8 (2.4)

Age at menopausea,b 48.9 (4.7) 47.9 (6.0)

Premenopausalc 15.0% 13.9%

Natural menopausec 41.8% 43.9%

Surgical menopausec 41.6% 40.5%

Alcohol beverages/day (1+)c 17.6% 11.2%

Any ®rst-degree family historyc 30.5% 21.5%

Nulliparousc 3.2% 3.7%

Age at ®rst birthd 24.1 (4.7) 23.5 (4.2)

a Mean and standard deviation.b Mean and standard deviation of age at menopause among

postmenopausal women who had a natural menopause or a bilateral

oophorectomy.c Percentage of cases or controls.d Mean and standard deviation of age at ®rst birth among parous

women.

Mammographic density and benign breast disease 105

density, and 9% had ³75% mammographic density. Theproportion of controls with ³50% breast density in-creased with the degree of proliferation, with 31.9% ofthose with nonproliferative conditions, 45.7% of thosewith proliferative disease without atypia, and 48.8% ofthose with atypical hyperplasia having ³50% mammo-graphic density. The mean percentage breast densityamong controls increased with the degree of prolifera-tion from 36.1% for those with nonproliferative condi-tions, 42.7% for proliferative disease without atypia, to48.2% with atypical hyperplasia. Women whose breasttissue was classi®ed as ``other'' had a mean percentagebreast density of 24.9%. The proportion of controlswith atypical hyperplasia increased from 6% to 20%as the percentage density rose from <10% to ³75%.Women with ``other'' types of benign conditions hadsigni®cantly lower percentage breast density (p £ 0.002),but the age and race adjusted percentage density meansacross the other benign histologic types were notstatistically signi®cantly di�erent (p ³ 0.157).The analyses of benign histology and breast cancer

risk are presented in Table 3. In analyses controllingonly for age and race, compared with those withnonproliferative disease, subjects with proliferative dis-ease without atypia had an OR of 1.3 (95% CI: 0.9±1.9)and those with atypical hyperplasia had an OR of 2.2(95% CI: 1.3±3.6). While the histologic classi®cationwas associated with the percentage mammographicdensity, additional adjustment for the percentage breastdensity had little e�ect on the ORs for proliferativelesions without atypia and for atypical hyperplasia(Table 3). Further adjustment for the breast cancer riskfactors of family history, drinking alcohol, nulliparityand age at ®rst birth, years of education, weight,menopause status, age at menopause, and use ofpostmenopausal hormones only slightly modi®ed themagnitude of the associations with benign histology.

Breast cancer risk also increased with the percentageof breast area with mammographic density among thesewomen with benign breast disease. Compared withwomen for whom less than 10% of the breast areaappears dense on the mammogram, the ORs were 2.0(95% CI: 1.2±3.4), 2.8 (95% CI: 1.6±4.7), and 4.1(95% CI: 2.1±7.8) for those with 10±49%, 50±74%,and ³75% of the breast area that appeared mammo-graphically dense, respectively (Table 4). Additionaladjustment for the type of benign histology slightlyreduced the e�ects associated with increased mammo-graphic density. Further adjustment for family history,drinking alcohol, nulliparity, age at ®rst birth, years ofeducation, and weight slightly increased the magnitudeof the associations with increased mammographicdensity such that women with ³75% mammographicdensity had an OR of 4.4 (95% CI: 2.1±9.0).To determine whether mammographic density was a

breast cancer risk factor across each histologic categoryof benign breast disease, and whether the category ofbenign breast disease was associated with risk acrosseach level of breast density, a cross-classi®ed variablewas created with women who had nonproliferativelesions and <50% breast density as the reference group(Table 5). Among women with nonproliferative benignbreast disease, risk increased almost six-fold (95% CI:1.8±18.6) for those with ³75% mammographic density.Breast cancer risk also increased with increasing per-centage mammographic density, among women withproliferative benign breast disease without atypia, withan OR of 1.6 (95% CI: 1.0±2.5) for <50% breastdensity, 2.8 (95% CI: 1.5±4.1) for 50±74% breast densi-ty, and 3.2 (95% CI: 1.6±6.6) for ³75% breast density.However, among women with atypical hyperplasia,there was no evidence of increased risk with greaterpercentage density.For women with <50% breast density, risk associat-

ed with the benign histology increased with the presenceand severity of epithelial proliferation. Among womenwith <50% breast density, those with proliferationwithout atypia had an OR of 1.6 (95% CI: 1.0±2.5), andthose with atypical hyperplasia had an OR of 4.1(95% CI: 2.1±8.0), compared with women with nonpro-liferative histology. In contrast, for women with ³75%breast density, risk decreased with the presence andseverity of proliferation. Among those women with³75% breast density, the OR was 5.8 (95% CI: 1.8±18.6) from the 11 cases and ®ve controls with nonpro-liferative histology, 3.2 (95% CI: 1.6±6.6) from the 29cases and 23 controls with proliferation without atypia,and 2.1 (95% CI: 0.6±7.0) from the six cases and sevencontrols with atypical hyperplasia, compared withthose with nonproliferative histology and <50% breast

Table 2. Joint distribution of breast histology and percentage breast

density among controls

Benign histology No.(%) of subjects by percentage

breast density

Total

<10% 10±49% 50±74% ³75%

Nonproliferative

disease

15 (16)a 49 (52) 25 (27) 5 (5) 94 [23]b

Proliferative disease

without atypia

29 (13) 92 (41) 79 (35) 23 (10) 223 [54]

Atypical hyperplasia 4 (9) 17 (42) 13 (32) 7 (17) 41 [10]

Other 19 (37) 24 (46) 9 (17) ± 52 [13]

Total 67 (16) 182 (44) 126 (31) 35 (9) 410

a Row percentages, in parentheses.b Column percentages, in square brackets.

106 C. Byrne et al.

density. The test of the multiplicative interaction termbetween the three categories of benign breast diseaseand the three levels of percentage breast density wassigni®cant at p = 0.002.

Discussion

While other studies have evaluated the associationbetween benign breast histology and mammographicfeatures [9±13], this is one of the ®rst studies to evaluatethe separate and joint e�ects of these factors on breast

cancer risk. In this study, benign breast disease histologyand measured percentage of the breast area withmammographic density appeared to be distinct breastcancer risk factors. The risk associated with benignbreast disease histology was not explained by the e�ectsof percentage breast density (Table 3), and the riskassociated with increased percentage breast density wasnot explained by the benign breast disease histology(Table 4). Furthermore, the breast cancer risks associ-ated with histologic types of benign breast disease andpercentage breast density were not explained by e�ectsof nulliparity and age at ®rst birth, ®rst-degree family

Table 5. Breast cancer risk by benign breast disease histology and percentage breast density

Benign histology category Percentage breast density

<50 50±74 ³75

Nonproliferative 1.0a ( ± ); 26/63c 2.5 (1.3±5.1)b; 24/23 5.8 (1.8±18.6); 11/5

Proliferative without atypia 1.6 (1.0±2.5); 88/119 2.5 (1.5±4.1); 78/79 3.2 (1.6±6.6); 29/23

Atypical hyperplasia 4.1 (2.1±8.0); 36/21 3.0 (1.3±7.0); 16/13 2.1 (0.6±7.0); 6/7

a Reference category. All results adjusted for age, race, family history, drinking alcohol, nulliparity and age at ®rst birth, years of education,

weight, menopause status, age at menopause, and use of postmenopausal hormones.b 95% con®dence interval.c Number of cases/number of controls in each category.

Table 3. Breast cancer risk by benign breast disease histology

Benign histology Cases Controls ORa

(95% CI)

ORb

(95% CI)

ORc

(95% CI)

Nonproliferative disease 62 94 1.0 1.0 1.0

Proliferative disease without atypia 198 223 1.3 (0.9±1.9) 1.3 (0.9±1.9) 1.2 (0.8±1.8)

Atypical hyperplasia 58 41 2.2 (1.3±3.6) 2.1 (1.3±3.6) 2.1 (1.2±3.6)

Other 29 52 0.8 (0.5±1.4) 0.9 (0.5±1.7) 0.8 (0.4±1.5)

a Logistic regression model includes age and race.b Logistic regression model includes percentage breast density, age, and race.c Logistic regression model includes percentage breast density, age, race, family history, drinking alcohol, nulliparity and age at ®rst birth,

years of education, weight, menopause status, age at menopause, and use of postmenopausal hormones. (Due to missing covariates, ®ve cases and

®ve controls were excluded from the full model analyses.)

Table 4. Breast cancer risk by percentage breast density

Percentage breast

density

Cases Controls ORa

(95% CI)

ORb

(95% CI)

ORc

(95% CI)

<10 26 67 1.0 1.0 1.0

10±49 143 182 2.0 (1.2±3.4) 1.9 (1.1±3.2) 2.0 (1.2±3.5)

50±74 127 126 2.8 (1.6±4.7) 2.6 (1.5±4.5) 3.0 (1.7±5.4)

³75 51 35 4.1 (2.1±7.8) 3.8 (2.0±7.2) 4.4 (2.1±9.0)

a Logistic regression model includes age and race.b Logistic regression model includes benign histology, age, and race.c Logistic regression model includes benign histology, age, race, family history, drinking alcohol, nulliparity and age at ®rst birth, years of

education, weight, menopause status, age at menopause, and use of postmenopausal hormones. (Due to missing covariates, ®ve cases and ®ve

controls were excluded from the full model analyses.)

Mammographic density and benign breast disease 107

history of breast cancer, years of education, alcohol use,body weight, menopause status, age at menopause, anduse of postmenopausal hormones (Tables 3 and 4).Having a high percentage breast density (³75%) wasassociated with almost a six-fold greater breast cancerrisk among women with nonproliferative benign breastdisease, and atypical hyperplasia was associated with agreater than four-fold rise in risk among women with<50% breast density (Table 5). However, the e�ects ofpercentage breast density seemed to modify the e�ectsassociated with atypical hyperplasia, and the e�ectsassociated with proliferative benign breast disease andatypical hyperplasia seemed to modify the e�ects asso-ciated with having ³75% breast density (Table 5).On a mammogram, epithelial and stromal tissue

appears dense compared to the translucent appearanceof fat. Thus, a mammogram provides an image of theentire breast from which the percentage of the breastwith mammographic density re¯ects the relative degreeof epithelial and stromal tissue compared with fat in thebreast. Several theories have been proposed to explainthe association between greater percentage breast den-sity and increased breast cancer risk. Trichopoulos andLippman [16] have suggested that the associationsupports the hypothesis of the importance of thenumber of epithelial cells in the etiology of breastcancer. In contrast, others have proposed that theincrease in stromal proliferation is likely to explain theassociation between percentage breast density andbreast cancer risk [7]. The results of this study cannotde®nitively di�erentiate between these hypotheses.However, since the risk associated with the mammo-graphic measure of percentage density was not explainedby the risk associated with the histologic measure ofepithelial proliferation, the results of this study supportthe theory that mammographic density may re¯ect anincreased proliferation of stromal tissue.Other studies have reported a similar increase in risk

with increased mammographic density [17±20]. Thisstudy was a subset of a larger case±control study nestedwithin the screening and follow-up phases of the BreastCancer Detection Demonstration Project (BCDDP) inwhich the initial screening mammogram was assessedfor 1880 cases and 2152 controls [6]. All cancersdiagnosed within one year of the initial mammogramwere excluded from this larger study, to eliminate thepotential bias in assessing the mammographic featuresfrom the ipsilateral breast. The relative risks for breastcancer associated with increased mammographic densityin the subset analyses presented in this paper were ofsimilar magnitude to those from the larger case±controlstudy, which indicates that this subset was not a biasedsample with regard to the mammographic e�ects. The

techniques for assessing the percentage of the breastwith dense mammographic appearance have variedacross studies [21]. Some studies used a visual estimateto categorize the percentage of the breast with mam-mographically dense appearance [17, 22], while othersmeasured the area of the breast and the area of densityto calculate the percentage of the breast with mammo-graphic density with a manual planimeter [18, 19], acomputerized planimeter [6], or a digitized computerimage [5, 20]. While still requiring an element ofsubjective evaluation, the computerized planimetermeasurement of the marked areas used for this substudyprovided a continuous measure of percentage breastdensity and was shown in the larger study to have a highintra- and inter-observer correlation coe�cient. Sincethe observers were blinded to the subsequent case±control status of the subjects, any misclassi®cation ofthe measure would likely have biased e�ects towards thenull.Proliferative epithelial lesions, particularly atypical

hyperplasia, have been associated with increased breastcancer risk in a number of studies [15, 23, 24]. From astudy of over 3000 biopsies with up to 24 years offollow-up, Dupont and Page reported that women withproliferative disease without atypia had a relative risk(RR) of 1.9 (95% CI: 1.2±2.9), and those with atypicalhyperplasia had a RR of 5.3 (95% CI: 3.1±8.8) com-pared with those who had nonproliferative disease [25].Similarly, Carter et al. reported that women withproliferative disease without atypia were at moderatelyincreased risk (RR = 1.9, 95% CI: 1.5±2.4), and wom-en with atypical hyperplasia were at the highest breastcancer risk (RR = 3.0, 95% CI: 2.1±4.1) [15]. This largeprospective study of 16,692 women who had a surgicalbreast biopsy diagnosed as benign during the BCDDPscreening phase, conducted by Carter et al., was thesource population for the nested case±control studypresented in this paper [15].While the cases and controls for the study presented in

this paper were sampled from the women in theprospective analyses, the magnitude of the e�ectsassociated with proliferative lesions and with atypicalhyperplasia in this case±control study were slightlylower than that reported by Carter et al. [15] for thispopulation (OR = 2.2 [95% CI: 1.3±3.6] in this studyvs. RR = 3.0, [95% CI: 2.1±4.1] in the previous study).There are several plausible reasons for these slightdi�erences. Due to the small numbers of subjects in thecase±control study, sampling variation may explain thedi�erences, especially since our 95% con®dence intervalsinclude the original e�ects. Alternatively, it is possiblethat the association with benign breast histology di�eredamong women for whom mammographic information

108 C. Byrne et al.

was available. This latter reason seems unlikely sinceanalyses conducted among the eligible cases and con-trols for whom mammographic data were not availablerevealed relative risks of 2.2 (95% CI: 1.1±4.1) forproliferative lesions without atypia and 2.2 (95% CI:0.9±5.5) for atypical hyperplasia. The third and mostlikely explanation for the slight di�erences in magnitudeof e�ects is the di�erence in time between the biopsy andbreast cancer in the two analyses. The large prospectiveanalysis evaluated the cancers diagnosed startingsix months after the biopsy (diagnosed between August1973 and February 1986). In the present nested case±control substudy, only breast cancer cases diagnosedafter the last BCDDP screening visit (in 1980) wereincluded. Thus, as the cohort aged, these analysesincluded a higher proportion of postmenopausal breastcancer cases and fewer premenopausal cases amongwhom atypical hyperplasia has a stronger impact. If theassociation between proliferative lesions and breastcancer risk decreases with time, as suggested by Dupontand Page [26], then excluding the cases diagnosed duringthe earlier screening phase would likely explain thelower relative risks in the case±control analysis.From this study, the joint e�ects of benign breast

disease histology and increased breast density were lessthan expected (p = 0.002), given the separate e�ects ofbenign breast disease histology among women with<50% breast density and the e�ects of increasedpercentage breast density for those with nonproliferativebenign breast disease. While this e�ect modi®cation maybe a spurious ®nding due to the small numbersevaluated in this study, even the upper level of the widecon®dence levels (0.7±8.9) for those with both factorsindicates risk no greater than additive for the combinede�ects of high breast density and atypical hyperplasia. Abetter understanding of the joint e�ects of benign breastdisease histology and percentage mammographic densi-ty will be an important step towards identifying whichwomen with proliferative benign breast disease are atparticularly high risk.

Acknowledgements

The authors thank Stuart Schnitt, MD and MarkSherman, MD for their helpful comments on earlierdrafts of this paper.

References

1. American Cancer Society (1995) Cancer Facts and Figures ± 1995.

Atlanta, GA.

2. Bodian CA (1993) Benign breast diseases, carcinoma in situ, and

breast cancer risk. Epidemiol Rev 15: 177±187.

3. Saftlas AF, Szklo M (1987) Mammographic parenchymal patterns

and breast cancer risk. Epidemiol Rev 9: 146±174.

4. Warner E, Lockwood G, Tritchler D, Boyd NF (1992) The risk of

breast cancer associated with mammographic parenchymal pat-

terns: a meta-analysis of the published literature to examine the

e�ects of method of classi®cation. Cancer Detect Prev 16: 67±72.

5. Boyd NF, Byng JW, Jong RA, et al. (1995) Quantitative classi-

®cations of mammographic densities and breast cancer risk: results

from the Canadian National Breast Screening Study. J Natl Cancer

Inst 87: 670±675.

6. Byrne C, Schairer C, Wolfe J, et al. (1995) Mammographic

features and breast cancer risk: e�ects with time, age and

menopause status. J Natl Cancer Inst 87: 1622±1629.

7. Oza AM, Boyd NF (1993) Mammographic parenchymal patterns:

a marker of breast cancer risk. Epidemiol Rev 15: 196±208.

8. Cook MG, Rohan TE (1985) The pathoepidemiology of benign

proliferative epithelial disorders of the female breast. J Pathol 146:

1±15.

9. Ingleby H, Gershon-Cohen J (1960) Comparative Anatomy,

Pathology, and Roentgenology of the Breast. Philadelphia, PA:

University of Philadelphia Press.

10. Wellings SR, Wolfe JN (1978) Correlative studies of the histo-

logical and radiographic appearance of the breast parenchyma.

Radiology 129: 299±306.

11. Bright RA, Morrison AS, Brisson J, et al. (1988) Relationship

between mammographic and histologic features of breast tissue in

women with benign biopsies. Cancer 61: 266±271.

12. Bartow SA, Pathak DR, Mettler FA (1990) Radiographic micro-

calci®cations and parenchymal patterns as indicators of histologic

``high-risk'' benign breast disease. Cancer 66: 1721±1725.

13. Boyd NF, Jensen HM, Cooke G, Han HL (1992) Relationship

between mammographic and histologic risk factors for breast

cancer. J Natl Cancer Inst 84: 1170±1179.

14. Boyd NF, Jensen HM, Cooke G, Han HL, Lockwood GA in

collaboration with the Reference Pathologists of the Canadian

National Breast Screening Study and Miller AB (2000). Mammo-

graphic densities and the prevalence and incidence of histological

types of benign breast disease. Eur J Cancer Prev 9: 15±24.

15. Carter CL, Corle DK, Micozzi MS, Schatzkin A, Taylor PR

(1988) A prospective study of the development of breast cancer in

16,692 women with benign breast disease. Am J Epidemiol 128:

467±477.

16. Trichopoulos D, Lippman RD (1992) Mammary gland mass and

breast cancer risk. Epidemiology 3: 523±526.

17. Brisson J, Merletti F, Sadowsky NL, Twaddle JA, Morrison AS,

Cole P (1982) Mammographic features of the breast and breast

cancer risk. Am J Epidemiol 115: 428±437.

18. Wolfe JN, Saftlas AF, Salane M (1987) Mammographic paren-

chymal patterns and quantitative evaluation of mammographic

densities: a case±control study. Am J Radiol 148: 1087±1092.

19. Saftlas AF, Hoover RN, Brinton LA, et al. (1991) Mammographic

densities and risk of breast cancer. Cancer 67: 2833±2838.

20. Boyd NF, Byng JW, Jong RA, et al. (1995) Quantitative classi-

®cations of mammographic densities and breast cancer risk: results

from the Canadian National Breast Screening Study. J Natl Cancer

Inst 87: 670±675.

21. Lee-Han H, Cooke G, Boyd NF (1995) Quantitative evaluation of

mammographic densities: a comparison of methods of assessment.

Eur J Cancer Prev 4: 285±292.

22. Boyd NF, O'Sullivan B, Campbell JE, et al. (1982) Mammograph-

ic signs as risk factors for breast cancer. Br J Cancer 45: 185±193.

Mammographic density and benign breast disease 109

23. Page DL, Dupont WD, Rogers LW, Rados MS (1985) Atypical

hyperplastic lesions of the female breast. Cancer 55: 2698±2708.

24. London SJ, Connolly JL, Schnitt SJ, Colditz GA (1992)

A prospective study of benign breast disease and the risk of breast

cancer. JAMA 267: 941±944.

25. Dupont WD, Page DL (1985) Risk factors for breast cancer in

womenwith proliferative breast disease.NEngl JMed 312: 146±151.

26. Dupont WD, Page DL (1989) Relative risk of breast cancer varies

with time since diagnosis of atypical hyperplasia. Hum Pathol 20:

723±725.

110 C. Byrne et al.